A Study on the Effect of Drift Factor on Feature Optimization in Electronic Nose Detection

Abstract

As an important instrument for olfactory detection in non-destructive testing (NDT), the electronic nose plays an important role in simulating olfactory detection. However, its performance is often affected by drift phenomena, including changes in sensor performance caused by environmental factors and olfactory fatigue due to long-term use. Although the effect of drift on the performance of electronic noses is widely recognized, there is still a relative lack of research on how drift affects feature optimization. This study presents a novel idea that drift factors not only affect the direct readings of the sensor but may also have a profound effect on the feature optimization process and hence the compensation of the electronic nose. To explore this concept, we chose temperature and humidity, the two most common environmental drift factors, for our experimental study. In our study, we verified the impact of drift factors on feature optimization and found a positive correlation between the concentration of sensor scores and the correct classification rate. Moreover, we adopted an innovative quadratic feature optimization method, which aims to reduce the influence of the drift factor on the feature optimization process and thus improve the drift resistance of the electronic nose. In our experiments, we found that the unweighted quadratic feature optimization method performs best in reducing the drift effect. After the optimization process, the recognition rate reaches 100% in the training set and 96% in the test set, which indicates that the electronic nose has improved significantly in terms of drift resistance. In summary, this study explores the effect of drift factors on feature optimization and proposes an effective treatment method, which provides a reference direction for the development and optimization of electronic nose technology.

Keywords

• e-nose
• drift compensation
• feature optimization
• random forest
• artificial neural networks